Functions for performing experimental comparisons of algorithms
using adequate sample sizes for power and accuracy. Implements the
methodology originally presented in Campelo and Takahashi (2019)
Felipe Campelo ([email protected]) and Fernanda Takahashi ([email protected])
Operations Research and Complex Systems Laboratory - ORCS Lab
Universidade Federal de Minas Gerais
Belo Horizonte, Brazil
Implementation of R package CAISEr, with routines for automatically determining the sample size needed for performing comparative experiments with algorithms.
To install the most up-to-date version directly from Github, simply type:
library(devtools)
devtools::install_github("fcampelo/CAISEr")
The most recent CRAN release of the package is also available for installation directly from the R prompt, using:
install.packages("CAISEr")
For instructions and examples of use, please take a look at the vignette
Adapting Algorithms for CAISEr, and at the package documentation, particularly
functions run_experiment()
and run_nreps2()
.
Please send any bug reports, questions, suggestions, chocolate (to Fernanda) or beers (to Felipe - we can always hope!) directly to the package authors listed at the top of this document.
Cheers,
Felipe
calc_se()
that resulted in NaN
if two vectors with the
same sample mean and same sample variance were passed as arguments.consolidate.partial.results()
)run_experiment()
can now be run in parallel using multiple cores.run_experiment()
and calc_nreps2()
can now save results to files.run_experiment()
now forces the use of all available instances if power >= 1
.CAISErPowercurve
objects.calc_power_curve()
to determine the range of effect sizes to consider.calc_nreps2()